Chapter 4 Do Multiple Credit Ratings Facilitate the Going Public Process?
4.4 Data and Sample
4.4.3 Sample identification and descriptive statistics
Table 1 (Panel A), offers a preliminary portrayal of our total sample (N=2,541) in comparison with sub-samples of enterprises with (N=252) and without (N=2289) acquisitions of credit ratings. The time span from the 1st of January, 1997 to the 31th of December 2014 includes IPO deals of companies that secured one or multiple ratings at least one year prior to the offering. This consortium illustrates interesting fluctuations in the number of awarded ratings. In particular, for the year 1998 data divulge a sufficient amount of rated firms (25) which is partly attributable to the record high new listing activity that begun with the ‘dot.com’ period that picked on March 2000 (Aggarwal, 2002).
The burst of this bubble at the end of 2000, coupled with major corporate failures and the subsequent economic slow-down of the U.S. economy, reduced the number of rated corporations almost by one-third in the next three years (9 ratings were identified in 2003). Then, rating acquisition increased again until the credit crunch of 2007-2008 (e.g. we report only one rated IPO deal in 2008). The CRA industry recovers the lost ground towards the end of our sample period (13 ratings in 2013). Hence, there is non-trivial verification that the frequency of credit rating possession by potential issuers strongly associates to the overall state of the economy.
Additionally, the right hand side of Panel A presents information on the allocation of ratings across years as well as the three leading U.S. CRAs under investigation. Consistently, we observe that the most preferred agency is Standard & Poor’s with a total of 111 awarded evaluations, followed by Moody’s with 86 and lastly by Fitch with 55. The three CRAs combined have given a rating to the 9.92% of the total sample. Panel B,
83
further analyzes the structure of credit ratings via exemplifying their assigned level. Interestingly, we notice that the bulk of rated companies range between the BBB+ and B- levels for Standard & Poor’s and Fitch and between Baa1 and Ba3 for Moody’s. Essentially, IPO deals concentrate around the boarder-line of lower medium investment grade and non investment speculative grades.
Table 2 supplies descriptive statistics for the complete sample as well as for the rated and nonrated new offerings. Appendix A contains detailed definitions of all the employed variables. Panel A puts forward four preliminary indications which support our main research hypothesis that rated firms incur less money left on the table. First, IPOs securing credit ratings document an average underpricing of a modest 18%. This results in an adequate 9 percentage point difference as opposed to the 27% first day return of their non rated counterparts. Second, repression of upward filling price revisions during the book-building process averts the need to bump up underpricing as a means of compensation to informed investors in order for them to unveil proprietary information (Hanley (1993), Loughran and Ritter (2002)). Therefore, in the credit rating sample exclusively, the average value of price revisions attains a minus sign (-2%). Third, Tobin’s Q as an effective proxy of a company’s competitive advantage (refer to Chung and Pruitt, 1995) also demonstrates inaugural evidence which sustain the notion that CRAs’ evaluations positively relate to distinguished short run IPO performance. This is evident at the mean Tobin’s Q ratio of rated new offerings which exceeds that of non rated issues by 27%. Lastly, the same conclusion holds true for our investor valuation estimator that appears 29% larger for firms that obtain credit ratings. Noticeably, the mean differences of all variables included in Panel A are statistically significant at the 5% and 1% level accordingly.
Panel B illustrates the IPO specific characteristics employed as control variables in all specification models. Proportionally, rated corporations are sufficiently larger than the non rated ones as surrogated by the mean of gross proceeds which amounts to almost $346 million raised by the former and $91 million by the latter IPO deals. This trend is also evident when average net sales are utilized as an alternative way to proxy for size while new listings with credit ratings take on less leverage.
On top of these stronger fundamentals, firms with credit ratings have longer time span of operations with an average age of about 26 years; which is almost 13 years older
84
than the average of the non rated sample. In line with their overall displayed quality, IPOs that maintain CRA evaluations usually rely on prestigious auditors and underwriters to facilitate the going public process and are not likely to seek venture capital backing. Additionally, they issue lower amount of primary shares while their presence is not eminent in NASDAQ which is a stock exchange that associates with more speculative listings (Lowry and Shu, 2002). Lastly, we observe no significant difference in the percent of shares overhang among new offerings with and without credit ratings.
4.5 Methodology
To fully capture the effects of credit ratings on IPO pricing we specify the following model:
Yi = α +β Xj + γ CR + ε (1)
Where Yi is the level of initial returns (or degree of price revision), Xj represents a vector of exogenous IPO relevant characteristics, CR enters the equation as a binary variable that is equal to unity when the firm secures one (or multiple) rating(s) and ‘ε’ stands for the disturbance term.
Initially, we conduct our analysis in a multivariate OLS regression setting. In order for coefficients to be unbiased, the estimate ‘γ’ of our main independent variable needs to be free from feedback effects and thus uncorrelated with ‘ε’ (Cov(CR,ε)=0). However, someone could argue that the acquisition of credit ratings is least partially decided by the firm’s management. It is plausible to assume that any company will seek CRAs evaluations if benefits, namely expectation of superior first trading day performance, outweigh the price required by the rating agencies. In this case, endogeneity and self selection bias could produce unreliable results.
Heckman (1979) argues that endogenous selection is very similar in nature with the omitted variables problem and proposes a two stage procedure to cope with it. Following this process we define a first stage regression that estimates the probability of a firm securing a rating. Specifically, we model this selection equation as:
85
CRi*= ω Wi + μ (2)
Where: 𝐶𝑅𝑖 = {1, 𝑖𝑓 𝐶𝑅𝑖 ∗> 0 0, 𝑖𝑓 𝐶𝑅𝑖∗≤ 0
In equation (2) CRi* is a latent variable, W a set of quantifiable determinants of CR, ω a vector of coefficients to be estimated and μ is the error term. Among the variables included in ‘W’ some could exert influence on IPO pricing and may be part of ‘ε’ in equation (1). Additionally, certain information that influence the company’s choice to seek ratings like R&D plans cannot be precisely measured and thus are included in ‘μ’. Correlation between the two error terms, verifies the existence of endogenous selection.
Following the intuition of Cohen (2003) and An and Chan (2008) who also deal in their research with endogenous dichotomous variables, we demonstrate our attempt to correct for self selection bias in the following augmented model:
E[Unerpricing or Price Revisions | CR = 1] =β′Χ+γ+ E [ε| CR = 1] =β′Χ+γ+ρσεφ (ω
′W)
Φ(ω′W) (3)
Similarly, the model for non-rated IPOs is:
E[Unerpricing or Price Revisions | CR = 0] = β′Χ+ ρσε1−−φ (ω′W)
Φ(ω′W) (4)
By deducting specifications (3) and (4) we derive the expected impact of credit rating(s) on the level of initial returns:
𝐸[Unerpricing or Price Rev. | 𝑃𝑀𝐶 = 1] − 𝐸[Unerpricing or Price Rev. | 𝑃𝑀𝐶 = 0]= 𝛾 +
𝜌 𝜎𝜀 𝜑 (𝜔 ′𝑊)
𝜑 (𝜔′𝑊)(1−𝛷 (𝜔′𝑊)) (5)
In which Φ and φ respectively, stand for the cumulative and density distribution function of the standard normal distribution.
Econometrically, equation (5) provides both the sign and scale of the effect of CRs on IPO pricing. This information is given via the coefficient ‘γ’ which corresponds to the OLS estimate in equation (1). However, now we can eliminate bias with the addition of the Inverse Mills Ratio (λ) that was missing from the initial multivariate analysis. The correction term conditional on rating existence, takes the following form:
86
𝜆 =𝛷 (𝜔𝜑 (𝜔′′𝑊))𝑊) if CR=1 or 𝜆 =1−𝛷 (𝜔−𝜑 (𝜔′′𝑊)𝑊)) if CR=0
This method describes the treatment effect model that we employ in order to tackle endogeneity. Additionally, we incorporate in our analysis a two stage instrumental variable (IV) approach in the spirit of Heckman (1978) and Wooldridge (2002). Under this framework we no longer have to assume normality in the distribution of residuals and thus the validity of results of our treatment analysis could be challenged. Essentially, in the 2SLS procedure the first step is a probit regression of the endogenous variable against the vector of all the available instruments that constitute ‘W’. In the second step, equation (1) is estimated with OLS while the dichotomous regressor CR is replaced by the fitted probabilities we obtained from the reduced form. The employment of predicted values as instruments is crucial for our analysis. Since the extant literature does not specifically dictate a set of parameters that should be included in equation (2), this methodology provides a degree of flexibility in the assortment of explanatory variables.